September 8, 2026
Your Customer Asked One Question. Google Asked Ten More. You Answered One.
AI search does not look up your keyword. It breaks the question into a pile of smaller ones you never ranked for, then cites whoever answered those. Here is what fan-out does to a local service business, with the real sub-questions pulled from Semrush, and how to stop losing a job you were never in the running for.
The question you ranked for is not the question that got asked
A homeowner types "kitchen remodel near me" into an AI search box. You rank for that. You have ranked for that since the Obama administration. You are feeling good.
The AI does not search for "kitchen remodel near me." It quietly rewrites the question into a stack of smaller ones. What does a kitchen remodel cost. How long does it take. Do I need a permit. Who is licensed in this county. Who has reviews that mention staying on budget. Then it goes and finds whoever answered each of those, staples the answers together, and hands the homeowner three names.
You answered one of those questions. The one you ranked for. The other nine went to a cabinet store in another state, a Reddit thread, and a guy with a YouTube channel. You did not lose the job. You were never in the running. (Which is worse, if you think about it, so don't.)
What fan-out actually is
Google calls it query fan-out, and it is the core move behind AI Mode and AI Overviews. One question in, many hidden searches out, one synthesized answer back. Semrush has a plain explainer on it, and the short version is that the model behaves less like a librarian and more like an intern who was told to make a spreadsheet.
How many hidden searches? Depends on who measured. The published counts for Google's AI Mode land around ten per question, give or take, and ChatGPT runs fewer. The exact number does not matter. What matters is that none of those hidden searches is the phrase you built your homepage around. They are the follow-up questions a nervous homeowner asks at the kitchen table.
So a page that wins the keyword and ignores the follow-ups is a page that wins a contest nobody is holding anymore.
The hidden questions, pulled from real search data
I did not want to guess what the sub-questions look like, so I pulled the question-format keywords around "kitchen remodel" from Semrush (US database, this week). Here is the top of the list, with monthly searches:
How much does a kitchen remodel cost: 4,400. How much should a 10x10 kitchen remodel cost: 1,000. How to remodel a kitchen: 880. How long does a kitchen remodel take: 720. How to plan a kitchen remodel: 480. Do you need a permit to remodel a kitchen: 320.
Read that list again as a contractor. Cost, timeline, permits, planning, small kitchens. Those are the questions you answer on the phone forty times a week. They are also the questions the AI is running behind the scenes when someone asks it who to hire. And your website answers exactly none of them, because your website is a photo of a faucet and a button that says Request a Quote.
Meanwhile the people who did answer them are a national cabinet retailer, a home improvement chain, and a Reddit thread titled something like "realistic kitchen remodel costs." None of them will show up to your customer's house. All of them got cited.
Answer the follow-ups on the page, not on the phone
The fix is not clever. Take the ten questions you answer on every sales call and write them down, in your own words, on the service page for that job. A cost range with the honest caveats. A timeline. Whether you pull the permit or they do. What is included and what is not. Financing, if you offer it. Put the same answers in your Google Business Profile Q&A, because the model reads that too.
This feels like giving away the sales call for free. It is. The sales call is the part the AI is doing now, with or without you. The only question is whether it does it with your numbers or a stranger's.
And no, a chatbot on your site does not count. The model is not going to click your chatbot. Nobody is going to click your chatbot. Your mother is not going to click your chatbot.
What I changed for the trades
My agency, Local Howl, does local SEO for service businesses, and fan-out changed how I write a service page. It used to be a headline, three paragraphs of adjectives, and a form. Now I build it around the follow-up questions: the cost range, the timeline, the permit answer, the what-is-included list, written the way the owner would say it on a call.
I pull the actual question list from search data first, the way I did above, instead of guessing which questions matter. Then I make sure the same answers live on the Google Business Profile, because a model corroborating a claim across two surfaces trusts it more than a claim it found once.
It is less glamorous than a redesign. It is also the difference between being one of the three names and being the fourth name nobody reads. (There is no fourth name. That is the point.)
Stop optimizing for the question and start answering the ones behind it
You need to know which hidden questions your category fans out into, and which ones you are currently invisible for. That is a data problem, and Semrush One is what I use for it: the Keyword Magic questions report gives you the sub-question list with volumes, and the AI Visibility Toolkit tells you whether the assistants are naming you or the cabinet store when those questions get asked.
If you want the wider picture, I wrote about why the assistant names some businesses and not others in the ChatGPT shortlist piece, and how to build the kind of entity a model can corroborate in the entity corroboration field guide.
One question in, ten hidden questions out. Answer them or watch a cabinet store in Ohio close your leads.